On-device
- Local inference for privacy and offline resilience
- Signed model distribution with integrity pinning
- Assist-only prompt contracts verified in CI
- No hidden network fallback
Loading...
AI Sense · bounded clinical assistance
MedXline AI can organize evidence, surface urgency, and prepare reviewable drafts. It cannot replace a licensed clinician, independently diagnose, prescribe, or make a final clinical decision.
Every assist identifies its source, limitations, reviewer, permitted next action, and decision record.
AI Sense
A plain-language concern—"chest tightness when climbing stairs, two weeks"—becomes a structured, labeled health summary. The output prepares a human; it never silently becomes clinical fact.
Plain-language concern capture
Structured health summary with timeline
Urgency signal and red-flag surfacing
Provider-specialty suggestion with rationale
Visit preparation and follow-up prompts
Consent-aware data handling
Decision log for every AI Sense output
Evaluation harness and guardrail policy in code
AI triage
Triage guides patients toward the right level of care and prepares providers with a structured intake summary. It is guidance, not diagnosis, and emergency signals bypass model conversation entirely.
Emergency escalation — red-flag symptoms route immediately to emergency instructions
Provider summary — concern, timeline, urgency, red flags, specialty, and stated limitations
Patient guidance — plain-language next steps with an explicit non-diagnostic boundary
Versioned evidence — models, prompts, and rules recorded with every interaction
Injection controls — fail-closed input and output screening before tools run
Human review — no output becomes clinical action without a licensed decision
Three layers · one governance model
Placement changes latency and privacy characteristics, never the requirement for explicit purpose, bounded tools, human review, audit, and an operational kill switch.
Inspect before adoption
Review the governance contract and evaluation posture, then test the system against your real policy and failure modes.